AI Incident Response Plan Template for Small Teams (2026)

A working AI incident response plan fits on two pages. This template gives you the five phases, role assignments, notification checklist, and timeline your team needs — ready to copy, fill in the blanks, and test in a 90-minute tabletop.

The 5-Phase AI Incident Response Plan

Phase 1: Detection and Reporting

Owner: Any team member who notices the incident
Timeline: Immediate — within 1 hour of discovery

What happens in Phase 1:

Someone notices something wrong and reports it. The barrier to reporting should be zero — teams that make incident reporting feel bureaucratic or punitive miss incidents.

Detection actions:

• Employee or monitoring system notices the incident
• Reporter contacts the Incident First Responder: [NAME] at [PHONE/SLACK]
• Reporter documents the initial observation (what they saw, when, on what system)
• Incident is logged in the incident tracker: [LOG LOCATION]

What counts as reportable:

• AI vendor notifies you of a breach affecting your data
• AI output causes or could cause harm to a user or third party
• Unauthorized AI tool discovered handling personal data
• AI system behaving unexpectedly in a way that affects users
• Data subject complaint about an AI decision

Phase 2: Triage and Severity Assessment

Owner: Incident First Responder
Timeline: Within 2 hours of detection

What happens in Phase 2:

The first responder assesses severity and decides whether to escalate to the Incident Commander. Not every report becomes a full incident — triage is the filter.

Severity levels:

Triage checklist:

• Review the incident report
• Assign severity level (Critical / High / Medium / Low)
• Notify Incident Commander if Critical or High: [NAME] at [PHONE/SLACK]
• Update the incident log with severity assessment

Phase 3: Containment and Evidence Preservation

Owner: Technical Lead
Timeline: Within 4 hours for Critical/High; within 24 hours for Medium

What happens in Phase 3:

Stop the bleeding. For a vendor breach: revoke API keys, disable the integration, and preserve a copy of the vendor notification. For an AI system misbehaving: disable the feature or route traffic away. For shadow AI: remove the tool and preserve the evidence of how it was used.

Containment actions (adapt to incident type):

Vendor breach:

• Preserve the vendor notification (screenshot or email copy)
• Disable or limit the affected integration if possible
• Request vendor incident report: what data was accessed, for how long, and by whom
• Preserve your DPA and any relevant vendor agreements

AI system misbehavior:

• Disable the AI feature or enable a fallback path
• Preserve logs of the AI outputs that caused the incident
• Document the scope: how many users affected, what outputs they received

Shadow AI discovery:

• Document what tool was in use, who used it, and what data was processed
• Remove the tool from production
• Preserve evidence (screenshots, access logs if available)

All incidents:

• Do not delete logs or evidence — preservation of records is required
• Record containment actions taken and timestamp each action

Phase 4: Notification

Owner: Incident Commander (decides); Legal Contact (advises on obligations)
Timeline: 72 hours is the default — do not wait for full investigation

Notification checklist:

Internal:

• CEO/executive team informed: [NAME] ☐ Done / Date: ______
• Legal counsel informed: [NAME OR FIRM] ☐ Done / Date: ______
• Board or investors (if material): ☐ Done / Date: ______

Regulatory (as applicable):

• GDPR supervisory authority (if personal data breach, 72-hour window): ☐ Done / Date: ______
• US state notification (if applicable, 30-72 hours): ☐ Done / Date: ______
• European AI Office (if you are a GPAI provider with systemic risk): ☐ Done / Date: ______

Vendor:

• Vendor notified of the incident (if the incident involves your data on their system): ☐ Done / Date: ______
• Vendor DPA rights invoked if applicable (audit right, termination right): ☐ Done / Date: ______

Affected individuals (if required by GDPR Article 34 or state law):

• Draft notification language reviewed by legal
• Individuals notified: ☐ Done / Date: ______

Phase 5: Post-Incident Review

Owner: Incident Commander
Timeline: Within 30 days of incident closure

What happens in Phase 5:

Every incident is a learning opportunity. The post-incident review answers three questions: What happened? Why did it happen? What changes prevent recurrence?

Post-incident review checklist:

• Root cause documented (not just symptoms — what failed in the process or system?)
• Timeline of the incident reconstructed
• What the response team did well
• What the response team would do differently
• Actions required: [list specific fixes, owners, deadlines]
• Incident response plan updated to reflect lessons learned
• Metrics updated: incident closed in the AI governance metrics dashboard

Roles and Contacts (Fill In)

Keep this contact sheet in a shared, offline-accessible location (printed copy or shared drive with offline access). Do not rely solely on a system that could be affected by the incident.

Testing the Plan: 90-Minute Tabletop Exercise

The most common failure in AI incident response: the plan exists but has never been tested. Gaps surface only during an actual incident — when there is no time to find them.

Run a tabletop every 6 months. Here is a ready-to-use scenario:

Scenario: Your company uses a third-party AI tool to process customer support tickets. The vendor sends an email: "We experienced a security incident affecting customer data processed by our platform between [date range]. Data potentially exposed includes names, email addresses, and support ticket content." Your company has 3,000 EU customers whose tickets were processed during this period.

Walk your team through Phases 1–5 using this scenario. Record gaps — roles that were unclear, contacts that were missing, notification obligations that surprised the team. Update the plan before you file it away.

For agentic AI systems specifically, see the AI security incident response playbook, which covers the additional containment and rollback steps when autonomous agents are involved.

The AI governance roles guide covers how to assign the Incident Commander and First Responder roles within your existing team structure.